# Claude Code Security Review — PR Audit Action

> Claude Code Security Reviewer is a GitHub Action that scans PR diffs for security issues and comments findings on the PR using a Claude API key.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

1. Add `.github/workflows/security.yml`:
   ```yaml
   name: Security Review
   permissions:
     pull-requests: write
     contents: read
   on:
     pull_request:
   jobs:
     security:
       runs-on: ubuntu-latest
       steps:
         - uses: actions/checkout@v4
           with:
             ref: ${{ github.event.pull_request.head.sha || github.sha }}
             fetch-depth: 2
         - uses: anthropics/claude-code-security-review@main
           with:
             comment-pr: true
             claude-api-key: ${{ secrets.CLAUDE_API_KEY }}
   ```
2. Add `CLAUDE_API_KEY` as a repo secret.
3. Open a PR and confirm the action posts findings as PR comments.

## Intro

Claude Code Security Reviewer is a GitHub Action that scans PR diffs for security issues and comments findings on the PR using a Claude API key.

- **Best for:** repos that want a diff-aware security pass on every PR before merging (especially backend/services)
- **Works with:** GitHub Actions, PR comment permissions, Claude API key secrets, trusted PR workflows
- **Setup time:** 8 minutes

## Practical Notes

- Diff-aware mode: README says it analyzes changed files for PRs (not full repo)
- Default model input in README references Opus 4.1 and a 20-minute ClaudeCode timeout (configurable)

## Using It Without Shooting Yourself in the Foot

AI security review is most useful when it’s **diff-scoped** and the repo has clear trust boundaries.

Recommended rollout:

- Enable it on internal PRs first (or require maintainer approval for external contributors) to reduce prompt-injection risk.
- Treat findings as a review aid, not an automatic block, until you calibrate false positives.
- Keep the action’s permissions minimal: it needs PR comment write access, not repo write access.

The README also documents customization via files (for example custom scan instructions and false-positive filtering). Adopt that once your team agrees on a “house style” for security comments.

### FAQ

**Q: Does it scan the whole repo?**
A: For PRs it focuses on changed files/diffs (per README).

**Q: Is it hardened against prompt injection?**
A: The README explicitly warns it is not; use trusted PR policies.

**Q: How do I tune false positives?**
A: Use the provided inputs for custom instructions / filtering files.

## Source & Thanks

> Source: https://github.com/anthropics/claude-code-security-review
> License: MIT
> GitHub stars: 4,568 · forks: 430

---

<!-- ZH -->

## 快速使用

1. 添加 `.github/workflows/security.yml`：
   ```yaml
   name: Security Review
   permissions:
     pull-requests: write
     contents: read
   on:
     pull_request:
   jobs:
     security:
       runs-on: ubuntu-latest
       steps:
         - uses: actions/checkout@v4
           with:
             ref: ${{ github.event.pull_request.head.sha || github.sha }}
             fetch-depth: 2
         - uses: anthropics/claude-code-security-review@main
           with:
             comment-pr: true
             claude-api-key: ${{ secrets.CLAUDE_API_KEY }}
   ```
2. 在仓库 Secrets 里添加 `CLAUDE_API_KEY`。
3. 打开 PR，确认 Action 在 PR 里发布安全发现（评论/建议）。

## 简介

Claude Code Security Reviewer 是一个 GitHub Action：只分析 PR 的变更文件，输出安全问题与修复建议并自动回评到 PR（需要 CLAUDE_API_KEY）。

- **适合谁：** 希望在合并前对每个 PR 做一次“只看 diff”的安全体检的仓库（尤其是后端/服务）
- **可搭配：** GitHub Actions、PR 评论权限、Claude API key secrets、可信 PR 流程
- **准备时间：** 8 分钟

## 实战建议

- 按 README：PR 场景只分析变更文件（diff-aware），避免全仓扫描带来的成本与噪音
- README 给出默认超时 20 分钟，并提供模型名参数（可按需求调整）

## 不翻车的使用方式

AI 安全审计最怕两件事：范围太大导致噪音爆炸，以及信任边界不清导致被 prompt injection 牵着走。

更稳的落地路径：

- 先在内部 PR 或“外部贡献需审批”模式下启用，降低被攻击面。
- 先把它当作 review 辅助，不要一开始就自动阻断合并；先校准误报率。
- 权限尽量最小化：写 PR 评论即可，不需要写仓库。

README 还支持用自定义指令文件来调优扫描与误报过滤；当团队确定“安全评论风格”后再逐步接入。

### FAQ

**会全仓扫描吗？**
答：按 README：PR 场景主要针对变更文件/差异，而不是全仓。

**能抵抗 prompt injection 吗？**
答：README 明确提示未做加固；建议只在可信 PR 策略下使用。

**怎么减少误报？**
答：使用 Action 提供的自定义扫描/过滤指令文件输入来调优。

## 来源与感谢

> Source: https://github.com/anthropics/claude-code-security-review
> License: MIT
> GitHub stars: 4,568 · forks: 430


---
Source: https://tokrepo.com/en/workflows/claude-code-security-review-pr-audit-action
Author: Script Depot